Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-8929

Chained replication is not working (non-transactional updates are replicated very slowly)

    XMLWordPrintable

Details

    • Bug
    • Status: Closed (View Workflow)
    • Major
    • Resolution: Done
    • 10.0.21
    • N/A
    • Replication
    • None
    • centos 6.6

    Description

      Hello,
      I began playing with MariaDB 10 and decided to check how it works with chained replication where 1 server replicates to 2nd and 2nd to 3rd.
      It's easily working with 5.5 but with 10.0.21 I've encountered a problem with 3rd server.

      The initial setup was:
      server 1 (5.5) -> server 2 (5.5) -> server 3 (5.5)

      After upgrade:
      server 1 (10.0) -> server 2 (10.0) -> server 3 (5.5)

      The next and final step was server 3 upgrade:
      server 1 (10.0) -> server 2 (10.0) -> server 3 (10.0)

      And here I got into troubles. Server 3 denied replicating. Counters worked fine (for different relay/master logs positions).
      Use of gtid on all servers was set to no. Actually I haven't touched anything related to gtid since wanted to keep using the old style replication.

      In the end I ended with 3rd server space problems. It kept downloading logs from master (server 2) without replaying it.

      So now I am back to server 1 (10.0) -> server 2 (10.0) -> server 3 (5.5) topology.

      In my tests (and future upgrades) I want to stay with classic/old replication style and enjoy MariaDB 10 parallel replication.
      So thought that without touching gtid settings it remains as is (like the server 1 and server 2 do).

      What is wrong with my chained replication setup that caused 3rd server to stop replicating?
      Is it something about default domain_id=0, something else or a bug that denied 3rd server ignore gtid related settings?

      Another thing I've paid my attention to - very high IO. Disk was at 100% usage (by mysql process) and always writing something.
      I thought that it was somehow caused by relay-logs massive download from master (server 2) and replaced disk with a faster one.
      I am using Amazon, so upgraded provisioned IOPS HDD , from 1000 IOPS to 2000 IOPS. It didn't help, and the the disk remained busy, same 100% and just doubled write speed.
      Can't say whether it's normal or not, when downgraded to 5.5 everything calmed down, no IO, no replication problems.

      Your help is much appreciated!

      Thanks,
      Alex

      Attachments

        Activity

          People

            knielsen Kristian Nielsen
            ShivaS Alex
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

            Dates

              Created:
              Updated:
              Resolved:

              Git Integration

                Error rendering 'com.xiplink.jira.git.jira_git_plugin:git-issue-webpanel'. Please contact your Jira administrators.